From: Matthew Brost <matthew.brost@intel.com>
To: Rodrigo Vivi <rodrigo.vivi@intel.com>
Cc: intel-xe@lists.freedesktop.org,
"Matthew Auld" <matthew.auld@intel.com>,
"Paulo Zanoni" <paulo.r.zanoni@intel.com>,
"Francois Dugast" <francois.dugast@intel.com>,
"Thomas Hellström" <thomas.hellstrom@linux.intel.com>
Subject: Re: [PATCH 12/12] drm/xe: Fix missing runtime outer protection for ggtt_remove_node
Date: Tue, 16 Jul 2024 17:32:00 +0000 [thread overview]
Message-ID: <ZpaukCbqvFKCyjDf@DUT025-TGLU.fm.intel.com> (raw)
In-Reply-To: <20240711171155.173717-12-rodrigo.vivi@intel.com>
On Thu, Jul 11, 2024 at 01:11:55PM -0400, Rodrigo Vivi wrote:
> Defer the ggtt node removal to a thread if runtime_pm is not active.
>
> The ggtt node removal can be called from multiple places, including
> places where we cannot protect with outer callers and places we are
> within other locks. So, try to grab the runtime reference if the
> device is already active, otherwise defer the removal to a separate
> thread from where we are sure we can wake the device up.
>
> v2: - use xe wq instead of system wq (Matt and CI)
> - Avoid GFP_KERNEL to be future proof since this removal can
> be called from outside our drivers and we don't want to block
> if atomic is needed. (Brost)
> v3: amend forgot chunk declaring xe_device.
> v4: Use a xe_ggtt_region to encapsulate the node and remova info,
> wihtout the need for any memory allocation at runtime.
> v5: Actually fill the delayed_removal.invalidate (Brost)
I think you lost this change in a rebase as delayed_removal.invalidate
doesn't appear to be set in this patch.
Matt
> v6: - Ensure that ggtt_region is not freed before work finishes (Auld)
> - Own wq to ensures that the queued works are flushed before
> ggtt_fini (Brost)
> v7: also free ggtt_region on early !bound return (Auld)
> v8: Address the null deref (CI)
> v9: Based on the new xe_ggtt_node for the proper care of the lifetime
> of the object.
>
> Cc: Matthew Auld <matthew.auld@intel.com>
> Cc: Paulo Zanoni <paulo.r.zanoni@intel.com>
> Cc: Francois Dugast <francois.dugast@intel.com>
> Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
> Cc: Matthew Brost <matthew.brost@intel.com>
> Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
> ---
> drivers/gpu/drm/xe/xe_ggtt.c | 106 ++++++++++++++++++-----------
> drivers/gpu/drm/xe/xe_ggtt_types.h | 12 ++++
> 2 files changed, 78 insertions(+), 40 deletions(-)
>
> diff --git a/drivers/gpu/drm/xe/xe_ggtt.c b/drivers/gpu/drm/xe/xe_ggtt.c
> index fe0bfd8c26cd..b57a71d54b1c 100644
> --- a/drivers/gpu/drm/xe/xe_ggtt.c
> +++ b/drivers/gpu/drm/xe/xe_ggtt.c
> @@ -161,6 +161,7 @@ static void ggtt_fini_early(struct drm_device *drm, void *arg)
> {
> struct xe_ggtt *ggtt = arg;
>
> + destroy_workqueue(ggtt->wq);
> mutex_destroy(&ggtt->lock);
> drm_mm_takedown(&ggtt->mm);
> }
> @@ -242,6 +243,8 @@ int xe_ggtt_init_early(struct xe_ggtt *ggtt)
> else
> ggtt->pt_ops = &xelp_pt_ops;
>
> + ggtt->wq = alloc_workqueue("xe-ggtt-wq", 0, 0);
> +
> drm_mm_init(&ggtt->mm, xe_wopcm_size(xe),
> ggtt->size - xe_wopcm_size(xe));
> mutex_init(&ggtt->lock);
> @@ -276,6 +279,67 @@ static void xe_ggtt_initial_clear(struct xe_ggtt *ggtt)
> mutex_unlock(&ggtt->lock);
> }
>
> +static void ggtt_node_remove(struct xe_ggtt *ggtt, struct xe_ggtt_node *node,
> + bool invalidate)
> +{
> + struct xe_device *xe = tile_to_xe(ggtt->tile);
> + bool bound;
> + int idx;
> +
> + if (!node || !node->ggtt)
> + return;
> +
> + bound = drm_dev_enter(&xe->drm, &idx);
> +
> + mutex_lock(&ggtt->lock);
> + if (bound)
> + xe_ggtt_clear(ggtt, node->base.start, node->base.size);
> + drm_mm_remove_node(&node->base);
> + node->base.size = 0;
> + mutex_unlock(&ggtt->lock);
> +
> + if (!bound)
> + goto free_node;
> +
> + if (invalidate)
> + xe_ggtt_invalidate(ggtt);
> +
> + drm_dev_exit(idx);
> +
> +free_node:
> + kfree(node);
> +}
> +
> +static void ggtt_node_remove_work_func(struct work_struct *work)
> +{
> + struct xe_ggtt_node *node = container_of(work, typeof(*node),
> + delayed_removal.work);
> + struct xe_device *xe = tile_to_xe(node->ggtt->tile);
> +
> + xe_pm_runtime_get(xe);
> + ggtt_node_remove(node->ggtt, node, node->delayed_removal.invalidate);
> + xe_pm_runtime_put(xe);
> +}
> +
> +/**
> + * xe_ggtt_node_remove - Remove a &xe_ggtt_node from the GGTT
> + * @ggtt: the &xe_ggtt where node will be removed
> + * @node: the &xe_ggtt_node to be removed
> + * @invalidate: if node needs invalidation upon removal
> + */
> +void xe_ggtt_node_remove(struct xe_ggtt *ggtt, struct xe_ggtt_node *node,
> + bool invalidate)
> +{
> + struct xe_device *xe = tile_to_xe(ggtt->tile);
> +
> + if (xe_pm_runtime_get_if_active(xe)) {
> + ggtt_node_remove(ggtt, node, invalidate);
> + xe_pm_runtime_put(xe);
> + } else {
> + queue_work(ggtt->wq, &node->delayed_removal.work);
> + }
> +}
> +
> /**
> * xe_ggtt_init - Regular non-early GGTT initialization
> * @ggtt: the &xe_ggtt to be initialized
> @@ -482,7 +546,9 @@ struct xe_ggtt_node *xe_ggtt_node_init(struct xe_ggtt *ggtt)
> if (!node)
> return ERR_PTR(-ENOMEM);
>
> + INIT_WORK(&node->delayed_removal.work, ggtt_node_remove_work_func);
> node->ggtt = ggtt;
> +
> return node;
> }
>
> @@ -499,46 +565,6 @@ void xe_ggtt_node_force_fini(struct xe_ggtt_node *node)
> kfree(node);
> }
>
> -/**
> - * xe_ggtt_node_remove - Remove a &xe_ggtt_node from the GGTT
> - * @ggtt: the &xe_ggtt where node will be removed
> - * @node: the &xe_ggtt_node to be removed
> - * @invalidate: if node needs invalidation upon removal
> - */
> -void xe_ggtt_node_remove(struct xe_ggtt *ggtt, struct xe_ggtt_node *node,
> - bool invalidate)
> -{
> - struct xe_device *xe = tile_to_xe(ggtt->tile);
> - bool bound;
> - int idx;
> -
> - if (!node || !node->ggtt)
> - return;
> -
> - bound = drm_dev_enter(&xe->drm, &idx);
> - if (bound)
> - xe_pm_runtime_get_noresume(xe);
> -
> - mutex_lock(&ggtt->lock);
> - if (bound)
> - xe_ggtt_clear(ggtt, node->base.start, node->base.size);
> - drm_mm_remove_node(&node->base);
> - node->base.size = 0;
> - mutex_unlock(&ggtt->lock);
> -
> - if (!bound)
> - goto free_node;
> -
> - if (invalidate)
> - xe_ggtt_invalidate(ggtt);
> -
> - xe_pm_runtime_put(xe);
> - drm_dev_exit(idx);
> -
> -free_node:
> - kfree(node);
> -}
> -
> /**
> * xe_ggtt_node_allocated - Check if node is allocated in GGTT
> * @node: the &xe_ggtt_node to be inspected
> diff --git a/drivers/gpu/drm/xe/xe_ggtt_types.h b/drivers/gpu/drm/xe/xe_ggtt_types.h
> index 60cbce3170d1..5ea12ea30a4d 100644
> --- a/drivers/gpu/drm/xe/xe_ggtt_types.h
> +++ b/drivers/gpu/drm/xe/xe_ggtt_types.h
> @@ -45,6 +45,8 @@ struct xe_ggtt {
> struct drm_mm mm;
> /** @access_count: counts GGTT writes */
> unsigned int access_count;
> + /** @wq: Dedicated unordered work queue to process node removals */
> + struct workqueue_struct *wq;
> };
>
> /**
> @@ -59,6 +61,16 @@ struct xe_ggtt_node {
> struct xe_ggtt *ggtt;
> /** @base: A drm_mm_node */
> struct drm_mm_node base;
> + /**
> + * @delayed_removal: Information for removal through work thread when
> + * device runtime_pm is suspended
> + */
> + struct {
> + /** @delayed_removal.work: The work struct for the delayed removal */
> + struct work_struct work;
> + /** @delayed_removal.invalidate: If it needs invalidation upon removal */
> + bool invalidate;
> + } delayed_removal;
> };
>
> /**
> --
> 2.45.2
>
next prev parent reply other threads:[~2024-07-16 17:33 UTC|newest]
Thread overview: 39+ messages / expand[flat|nested] mbox.gz Atom feed top
2024-07-11 17:11 [PATCH 01/12] drm/xe: Removed unused xe_ggtt_printk Rodrigo Vivi
2024-07-11 17:11 ` [PATCH 02/12] drm/xe: Introduce GGTT documentation Rodrigo Vivi
2024-07-11 19:41 ` Michal Wajdeczko
2024-07-11 17:11 ` [PATCH 03/12] drm/xe: Remove unnecessary drm_mm.h includes Rodrigo Vivi
2024-07-11 18:35 ` Cavitt, Jonathan
2024-07-11 17:11 ` [PATCH 04/12] drm/{i915, xe}: Avoid direct inspection of dpt_vma from outside dpt Rodrigo Vivi
2024-07-11 18:38 ` Cavitt, Jonathan
2024-07-11 17:11 ` [PATCH 05/12] drm/xe: Encapsulate drm_mm_node inside xe_ggtt_node Rodrigo Vivi
2024-07-11 19:51 ` Michal Wajdeczko
2024-07-16 17:16 ` Matthew Brost
2024-07-11 17:11 ` [PATCH 06/12] drm/xe: Rename xe_ggtt_node related functions Rodrigo Vivi
2024-07-16 17:24 ` Matthew Brost
2024-07-11 17:11 ` [PATCH 07/12] drm/xe: Limit drm_mm_node_allocated access to xe_ggtt_node Rodrigo Vivi
2024-07-17 23:27 ` Matthew Brost
2024-07-11 17:11 ` [PATCH 08/12] drm/xe: Introduce xe_ggtt_largest_hole Rodrigo Vivi
2024-07-11 20:00 ` Michal Wajdeczko
2024-08-09 21:33 ` Rodrigo Vivi
2024-07-11 17:11 ` [PATCH 09/12] drm/xe: Introduce xe_ggtt_print_holes Rodrigo Vivi
2024-07-11 18:43 ` Cavitt, Jonathan
2024-07-11 20:09 ` Michal Wajdeczko
2024-08-09 21:38 ` Rodrigo Vivi
2024-07-11 17:11 ` [PATCH 10/12] drm/xe: Rename xe_ggtt balloon functions to make the node clear Rodrigo Vivi
2024-07-11 20:17 ` Michal Wajdeczko
2024-08-09 21:42 ` Rodrigo Vivi
2024-07-11 17:11 ` [PATCH 11/12] drm/xe: Make xe_ggtt_node struct independent Rodrigo Vivi
2024-07-17 23:50 ` Matthew Brost
2024-07-11 17:11 ` [PATCH 12/12] drm/xe: Fix missing runtime outer protection for ggtt_remove_node Rodrigo Vivi
2024-07-16 17:32 ` Matthew Brost [this message]
2024-07-11 17:25 ` ✓ CI.Patch_applied: success for series starting with [01/12] drm/xe: Removed unused xe_ggtt_printk Patchwork
2024-07-11 17:26 ` ✓ CI.checkpatch: " Patchwork
2024-07-11 17:28 ` ✓ CI.KUnit: " Patchwork
2024-07-11 17:34 ` ✗ CI.Build: failure " Patchwork
-- strict thread matches above, loose matches on Subject: below --
2024-08-15 22:07 [PATCH 01/12] " Rodrigo Vivi
2024-08-15 22:07 ` [PATCH 12/12] drm/xe: Fix missing runtime outer protection for ggtt_remove_node Rodrigo Vivi
2024-08-16 15:02 [PATCH 01/12] drm/xe: Removed unused xe_ggtt_printk Rodrigo Vivi
2024-08-16 15:02 ` [PATCH 12/12] drm/xe: Fix missing runtime outer protection for ggtt_remove_node Rodrigo Vivi
2024-08-16 16:08 ` Lucas De Marchi
2024-08-17 10:14 ` Rodrigo Vivi
2024-08-17 10:35 [PATCH 01/12] drm/xe: Removed unused xe_ggtt_printk Rodrigo Vivi
2024-08-17 10:35 ` [PATCH 12/12] drm/xe: Fix missing runtime outer protection for ggtt_remove_node Rodrigo Vivi
2024-08-20 20:25 [PATCH 01/12] drm/xe: Removed unused xe_ggtt_printk Rodrigo Vivi
2024-08-20 20:25 ` [PATCH 12/12] drm/xe: Fix missing runtime outer protection for ggtt_remove_node Rodrigo Vivi
2024-08-21 19:38 [PATCH 01/12] drm/xe: Removed unused xe_ggtt_printk Rodrigo Vivi
2024-08-21 19:38 ` [PATCH 12/12] drm/xe: Fix missing runtime outer protection for ggtt_remove_node Rodrigo Vivi
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=ZpaukCbqvFKCyjDf@DUT025-TGLU.fm.intel.com \
--to=matthew.brost@intel.com \
--cc=francois.dugast@intel.com \
--cc=intel-xe@lists.freedesktop.org \
--cc=matthew.auld@intel.com \
--cc=paulo.r.zanoni@intel.com \
--cc=rodrigo.vivi@intel.com \
--cc=thomas.hellstrom@linux.intel.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox